⚡ 本页包含 AI 生成的分析内容，仅供参考
该论文提出了一种用于3D虚拟现实的张量显示处理器，通过算法-架构协同优化将焦堆栈转换为张量显示，降低了计算和硬件复杂度。在200MHz时钟频率下实现了161Mpixels/s的最大吞吐量（等效于HD分辨率下87fps），归一化能耗为0.9nJ/pixel，PSNR为23.7-24.8dB。
Abstract This work presents a 3D display processor that converts the focal stack to tensor display. Algorithm-architecture co-optimization is applied to reduce the computational complexity and hardware complexity. It achieves the maximum throughput of 161 Mpixels/s (equivalent to 87 fps at HD resolution) at a clock frequency of 200 MHz, with normalized energy consumption of 0.9nJ/pixel, and a PSNR of 23.7 to 24.8dB, The metaverse is progressively becoming technically feasible through integration of sensing technology, virtual reality (VR), and 3D displays [1,2], as illustrated in Fig. 24.3.1. For 3D displays, tensor display is a promising solution to creating 3D effects by showing the scenes on multiple image layers [3]. Given multiple light-field (LF) images (captured by an LF camera), the image layers can be generated through non-negative matrix factorization (NMF) [3,4]. This process is highly time-consuming because a large volume of LF data needs to
Ching-Yen Lee, Yueh-Feng Tsai, Ying-Sheng Lin, Chia-Hsiang Yang
National Taiwan University, Taipei, Taiwan